Minimum norm estimates...

[Robert Oostenveld]

Hi Suresh

On 13 May 2008, at 16:57, Suresh Muthukumaraswamy wrote:
>  I am trying to use sourceanalysis to create minimum norm
> estimates. The
> software runs fine but the resulting images I get back are completely
> dominated by the voxels at the edges of my grid near the sensors.
> So I am
> wondering If I have either made a mistake somewhere or if there are
> some
> parameters that can be tweaked to improve the reconstructions?

The standard MNE is non-regularised. Since superficial dipoles result
in a much stronger leadfield, the superficial sources are able to
explain your data with much "smaller" norm, hence the minimum norm
solution is very superficial. That is the depth bias inherent to MNE.

The minimum norm estimate implementation has the following options

%  'noisecov'         = Nchan x Nchan matrix with noise covariance
%  'sourcecov'        = Nsource x Nsource matrix with source
covariance (can be empty, the default will then be identity)
%  'lambda'           = scalar, regularisation parameter (can be
empty, it will then be estimated from snr)
%  'snr'              = scalar, signal to noise ratio
%  'reducerank'       = reduce the leadfield rank, can be 'no' or a
number (e.g. 2)
%  'normalize'        = normalize the leadfield
%  'normalizeparam'   = parameter for depth normalization (default =
0.5)

These can be passed in sourceanalysis by specifying e.g.
cfg.mne.normalize = 'yes'.

See also the documentation included in the private functions
minimumnormestimate and compute_leadfield.

My first guess at the solution to your problem would be normalize and
optionally the normalizeparam option. Depth regularization (or "depth
weighing") is described in various papers and I recall that the
normalizeparam comes from a paper by Wagner and/or Fuchs (Hamburg).

I hope this suggestion gets you started. Since we have not used MNE
ourselves sofar, I am curious to hear how it performs on your 275ch
MEG data and which tweaks are most optimal.

best regards,
Robert


> My data is 275Ch CTF data and for the purposes of trying out
> fieldtrip's MNE
> I am just using a simple VEP dataset, with the headmodel created by
> localSpheres. There is a simple field pattern in the data which I
> was hoping
> to reconstruct.
>   I have used the DICS (power) part of sourceanalysis before and
> get images
> that look pretty similar to SAM images made outside of fieldtrip,
> so I think
> my MRI/registration stuff in fieldtrip is ok.
>
> Thanks,
>   Suresh
>
> Here is my code
>
> Dataset = '/gpfs/home/sapsm7/Bristol/Day/DC/DC_VSM-
> VEF_20080409_01.ds';
> hdmFile = [Dataset '/default.hdm'] ;%assumes localSpheres has been run
> mrifile = '/gpfs/home/sapsm7/Bristol/Day/DC/DC080409.mri';
> shapefilename = '/gpfs/home/sapsm7/Bristol/Day/DC/DC080409.shape';
> OutImage = '/gpfs/home/sapsm7/Bristol/Day/DC/testminnorm';
>
>
> cfg = [];
> cfg.trialdef.prestim = 2; %How much of the prestim to load in (s)
> cfg.trialdef.poststim = 2;
> cfg.trialdef.eventtype = 'High';
> cfg.dataset = Dataset
> cfg.channel = {'MEG'}; %Right V5 and Left V5 sensitive
> cfg.bpfilter = [0.5 30];
> cfg.lpfilter = 30
> cfg.blc = 'yes';
> cfg.blcwindow = [-0.2 0];
> cfg = definetrial(cfg);
> Data = preprocessing(cfg);
>
> save DC_RawData
>
> load DC_RawData
>
> cfg = [];
> cfg.latency = [-0.2 1];
> DCERF = timelockanalysis(cfg, Data);
>
> cfg = [];
> cfg.baseline = [-0.2 0];
> DCERF = timelockbaseline(cfg, DCERF);
>
> %At this point the waveform plots and topos in fieldtrip look very
> nice
>
>
> [XMax, XMin, YMax, YMin, ZMax, ZMin] = shapefilemaxmin
> (shapefilename) ;
> %Extract the source reconstruction grid from the shapefile
> Resolution = 1;
> cfg = [];
> cfg.method = 'mne';
> cfg.hdmfile = hdmFile;
> cfg.grid.xgrid = XMin : Resolution : XMax;
> cfg.grid.ygrid = YMin : Resolution : YMax;
> cfg.grid.zgrid = ZMin : Resolution : ZMax;
> [sourcePre] = sourceanalysis(cfg, DCERF);
>
> sourcePre.avg.coll = mean(sourcePre.avg.pow(:,169:183), 2); %Grab a
> time
> window and put into a variable for sourceint
>
> cfg = [];
> cfg.downsample = 2;
> sourceInt = sourceinterpolate(cfg, sourcePre, mrifile);
>
>
> sourceInt.avg.coll = sourceInt.avg.coll * 1e10 %upscale to make it
> easy to
> visualise in mri3dX
>
> %Write the analyse files out
> DeleteAnalyseFiles(OutImage);
> cfg = [];
> cfg.parameter     = 'coll';
> cfg.filename      = OutImage;
> cfg.filetype      = 'spm';
> cfg.coordinates   = [];
> cfg.datatype = 'float';
> cfg.coordinates = 'ctf';
> volumewrite(cfg, sourceInt);
>
> ----------------------------------
> The aim of this list is to facilitate the discussion between users
> of the FieldTrip  toolbox, to share experiences and to discuss new
> ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/
> archives/fieldtrip.html and http://www.ru.nl/fcdonders/fieldtrip.
>

----------------------------------
The aim of this list is to facilitate the discussion between users of the FieldTrip  toolbox, to share experiences and to discuss new ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/archives/fieldtrip.html and http://www.ru.nl/fcdonders/fieldtrip.